In May 2023, the Semantic Kernel team put out a short post announcing two sessions at Microsoft Build, the first real stage appearance for what was then a fairly new open source SDK. The post itself was brief, just a couple of paragraphs pointing people to the sessions, but it marks a useful moment to look back on. Semantic Kernel went from a side project inside Microsoft to becoming one of the main building blocks for enterprise AI applications on Azure, and these two Build sessions were where that story started to go public.
What the original announcement said
The post confirmed that Semantic Kernel would be part of Build 2023, running from May 23rd to May 25th, and pointed attendees to two specific sessions. Both were framed around a simple idea: developers needed a new way to combine large language models with existing code, and Semantic Kernel was Microsoft’s answer to that problem at the time.
Session 1: Building AI solutions with Semantic Kernel
This session covered why the team built Semantic Kernel in the first place and framed it as requiring a different mindset from traditional application development. It also touched on how the SDK evolved alongside GPT-4’s rollout and introduced the plugin model, which let developers wrap native code and prompts behind a common interface that the model could call. The talk closed with early observations gathered from teams already experimenting with Semantic Kernel for copilot style features.
This was available to both in-person and online attendees, on Tuesday, May 23rd at 5:15 PM Pacific time. Making it available online was a deliberate choice, since the plugin model was still new enough that most developers outside Microsoft had not seen a working example of it yet.
Session 2: Building an AI Copilot with Semantic Kernel in the GPT-4 era
The second session went further into Microsoft’s broader plan for adding AI features using OpenAI and Azure OpenAI Service, with Semantic Kernel positioned as one piece of that toolchain rather than the whole answer. It walked through the SDK’s architecture and shared findings from internal hackathons where teams had used Semantic Kernel to prototype copilots, including which patterns worked and which ones did not scale well.
Unlike the first session, this one was restricted to in-person attendees only, on Wednesday, May 24th at 11:00 AM Pacific time. That restriction is a small but telling detail. Microsoft treated the architecture deep dive as content worth reserving for people physically at the conference, while the introductory session was made available more broadly.
Why this announcement mattered more than it looked
On the surface this was just a two paragraph event listing, but it captured a turning point. Before Build 2023, Semantic Kernel was mostly known through GitHub and a handful of blog posts. Getting stage time at Microsoft’s flagship developer conference, alongside a direct connection to GPT-4’s launch, signaled that Microsoft was betting on it as a supported path for building AI features in production, not just a research toy.
The mention of hackathon findings in Session 2 is also worth noting. It meant Microsoft already had internal production feedback before making the framework’s public pitch, which is a healthier way to introduce a new SDK than announcing it purely on paper. Teams evaluating a new framework today should look for the same signal: has anyone actually shipped something with it, or is it still theoretical.
Where Semantic Kernel stands today
Looking at this from 2026, it is worth being direct about how the framework has changed since that announcement. The Stepwise and Handlebars planners that were part of the early orchestration story have since been deprecated across the .NET, Python, and Java versions of Semantic Kernel, in favor of native function calling, which reasoning models now handle more reliably on their own. If you find older tutorials referencing planners, treat that code as outdated rather than a starting point.
More significantly, Microsoft has since merged Semantic Kernel’s orchestration and enterprise integration strengths with AutoGen’s multi-agent conversation capabilities into a single Microsoft Agent Framework, which went to public preview in October 2025 with general availability targeted for the first quarter of 2026. Semantic Kernel is not being killed off overnight, but new projects are steered toward Agent Framework, and Microsoft has published a migration guide for teams moving existing Semantic Kernel code over.
Practical guidance for teams evaluating this today
If you are maintaining an existing Semantic Kernel application, there is no urgent need to rip it out. The core plugin and function calling concepts carry over conceptually into Agent Framework, so the migration is more of a planned effort than a rewrite from scratch. Budget time to read Microsoft’s official migration guide before starting, since API names and package references have changed even where the underlying ideas have not.
For a brand new project, it makes more sense to start directly with Microsoft Agent Framework rather than Semantic Kernel, particularly if you expect to need multi-agent orchestration or native Model Context Protocol support, both of which are first class in the newer framework. Semantic Kernel still has value as a simpler single agent orchestration layer if your use case does not need that complexity, but be aware you are choosing a framework that is being consolidated rather than actively expanded.
Leave a Reply